`[GPRJ]` misses experiments its project holds¶

Archive: NCBI E-utilities (db=sra) Affected routes: bioproject->run:efetch_direct, bioproject->run:sra_be_direct_cgi Status: deterministic; nothing in the route recovers it. The union with the ELink and ENA routes does. Measured: 2026-10-26, over the 12,755 BioProjects of reprocessed-prj; reproduced the same day

Symptom¶

An ESearch of db=sra for PRJNA66167[GPRJ] finds 2,902 experiments, and their runs number 5,499. ENA, ELink and NCBI's own runinfo all put 5,910 runs in the project. The 411 missing runs are in experiments that are not indexed under the project's [GPRJ], although every record of them names it:

SRX135150[ACCN] finds the experiment.
SRX135150[ACCN] AND PRJNA66167[GPRJ] finds nothing.
runinfo?acc=SRR453077, a run of that experiment, gives BioProject as PRJNA66167.

The same holds for all 30 experiments of its study, SRP012040.

No error is given and the search is not rewritten. The query translation is PRJNA66167[GPRJ] as asked, and the answer is a smaller project.

The number¶

Over the 12,755 projects, the direct routes miss 1,100 runs in 2 projects that ELink returns and ENA lists under the same project:

Project	`[GPRJ]` finds	ELink finds	Missed by `[GPRJ]`, listed by ENA
PRJNA30709	7,665 runs	8,574	689
PRJNA66167	5,499 runs	5,910	411

A third difference, PRJNA1035929 (0 against 8), was timing: the project was empty to [GPRJ] when surveyed and answers 8 now.

Reproducer¶

E=https://eutils.ncbi.nlm.nih.gov/entrez/eutils
curl -s "$E/esearch.fcgi?db=sra&term=SRX135150%5BACCN%5D&retmode=json&retmax=0" | grep -o '"count":"[0-9]*"'
# "count":"1"
curl -s "$E/esearch.fcgi?db=sra&term=SRX135150%5BACCN%5D+AND+PRJNA66167%5BGPRJ%5D&retmode=json&retmax=0" | grep -o '"count":"[0-9]*"'
# "count":"0"
curl -s "https://trace.ncbi.nlm.nih.gov/Traces/sra-db-be/runinfo?acc=SRR453077" | cut -d, -f1,22 | head -2
# the run's BioProject column: PRJNA66167

Checked on 2026-10-26.

What it is not¶

Not a fall-through to [All Fields]. That was this slug's predecessor, gprj-falls-through-to-all-fields, now retired:
- every [GPRJ] search checked kept its field;
- the 22 projects [GPRJ] still finds empty answer phrasesnotfound, not a broader search;
- so does a project accession that does not exist.
Not ELink's error. ELink returns too much, in other projects. Here [GPRJ] returns too little, in this one.

Workaround¶

None inside the route. The resolver's union mode asks the ELink and ENA routes too, and both return these runs, so the default output holds them. A caller using Mode.FIRST gets ENA first, which holds them.

Affected accessions¶

python3 -c "
import csv, gzip, json
read = lambda r: {x['accession']: set(json.loads(x['results'] or '[]')) for x in csv.DictReader(gzip.open(f'data/surveys/reprocessed-prj/bioproject-to-run_{r}.csv.gz', 'rt'))}
d, e, n = read('efetch_direct'), read('efetch_elink'), read('ena_filereport')
print({a: len((e[a] - d[a]) & n.get(a, set())) for a in d if (e[a] - d[a]) & n.get(a, set())})"

Regression test¶

tests/pathologies/test_gprj_misses_experiments_of_the_project.py asserts that SRX135150 is still outside PRJNA66167[GPRJ] while its run's runinfo still names PRJNA66167.

[GPRJ] misses experiments its project holds¶